提交 · 2f01f19daed9702aa5bedab631e8b4028967f6f7 · testgroup / pytensor

09 7月, 2016 40 次提交
- Fixed concatenation error · 2f01f19d
  由 sentient07 提交于 6月 17, 2016
  
  2f01f19d
- Rework the subtensor lifter. If it have only 1 clients, don't move it to the… · b1b05e86
  由 Frederic Bastien 提交于 6月 16, 2016
```
Rework the subtensor lifter. If it have only 1 clients, don't move it to the GPU. Simpler condition that do what we want.
```
  b1b05e86
- Special Alloc and AllocEmpty to move them to the GPU in one pass. · fbc384cf
  由 Frederic Bastien 提交于 6月 16, 2016
  
  fbc384cf
- small code simplification · 12fa1c5b
  由 Frederic Bastien 提交于 6月 16, 2016
  
  12fa1c5b
- Move mrg_uniform to the GPU · 0bebe810
  由 Frederic Bastien 提交于 6月 16, 2016
  
  0bebe810
- Enable the condition to move · c9752903
  由 Frederic Bastien 提交于 6月 16, 2016
  
  c9752903
- Use the good name, but it isn't used, just clarification. · 1bc89154
  由 Frederic Bastien 提交于 6月 16, 2016
  
  1bc89154
- Cache GpuDnnConvDesc · 78aa9113
  由 Frederic Bastien 提交于 6月 16, 2016
  
  78aa9113
- Printing the nodes that aren't transferred · e6b9c271
  由 sentient07 提交于 6月 16, 2016
  
  e6b9c271
- Don't move to the GPU scalar int and make GraphToGPU support target. · 658563c8
  由 Frederic Bastien 提交于 6月 15, 2016
  
  658563c8
- Simpler dnn conv op graph · 31dcc3f7
  由 Frederic Bastien 提交于 6月 15, 2016
  
  31dcc3f7
- Build smaller conv shape graph. · 5047b874
  由 Frederic Bastien 提交于 6月 15, 2016
  
  5047b874
- Don't register ShapeOptimizer in fast_compile. local_shape_to_shape_i isn't… · 0691fb01
  由 Frederic Bastien 提交于 6月 15, 2016
```
Don't register ShapeOptimizer in fast_compile. local_shape_to_shape_i isn't registered, so it won't be used.
```
  0691fb01
- Get the shape of the node closer to inputs. · 3ff56c1d
  由 Frederic Bastien 提交于 6月 15, 2016
  
  3ff56c1d
- Small speed up. · c16c66eb
  由 Frederic Bastien 提交于 6月 15, 2016
  
  c16c66eb
- A small optimization · 52baaf7d
  由 sentient07 提交于 6月 15, 2016
  
  52baaf7d
- Added outputs argument · e8017096
  由 sentient07 提交于 6月 15, 2016
  
  e8017096
- Implement the complicated op_lifter condition correctly for GraphToGPU. · 3510323b
  由 Frederic Bastien 提交于 6月 14, 2016
  
  3510323b
- Fix cycle with GraphToGPU. · 7d303ff1
  由 Frederic Bastien 提交于 6月 14, 2016
  
  7d303ff1
- In the new back-end if config.dnn.enabled is False, don't try to compile dnn version. · 5c8e2cbf
  由 Frederic Bastien 提交于 6月 14, 2016
  
  5c8e2cbf
- Addressed comments · c2a6a650
  由 sentient07 提交于 6月 14, 2016
  
  c2a6a650
- Fixed IfElse opt and few subtensor tests · c8ba3f55
  由 sentient07 提交于 6月 14, 2016
  
  c8ba3f55
- Cleaned up and fixed pep8 · d0dfb0be
  由 sentient07 提交于 6月 14, 2016
  
  d0dfb0be
- Moved the time computation out of loop · c3e8f153
  由 sentient07 提交于 6月 14, 2016
  
  c3e8f153
- moved the node_created out of the loop · 5dac6c2d
  由 sentient07 提交于 6月 13, 2016
  
  5dac6c2d
- Print in the profiling result more compactly · fab5ab7b
  由 sentient07 提交于 6月 13, 2016
  
  fab5ab7b
- Cached three more Ops · d46b621f
  由 sentient07 提交于 6月 13, 2016
  
  d46b621f
- Made few opts accept outputs · d4625800
  由 sentient07 提交于 6月 13, 2016
  
  d4625800
- Removed certain params from print_profile · 1b1a8505
  由 sentient07 提交于 6月 11, 2016
  
  1b1a8505
- Added print_profile2 · b5772416
  由 sentient07 提交于 6月 10, 2016
  
  b5772416
- Added print_profile to GraphtoGPU · c396d611
  由 sentient07 提交于 6月 10, 2016
  
  c396d611
- Reshape crashfix and added caching at other places · de536bd5
  由 sentient07 提交于 6月 09, 2016
  
  de536bd5
- Cache GpuAllocEmpty · e01db583
  由 sentient07 提交于 6月 09, 2016
  
  e01db583
- Added new optimizer to local_abstractconv_cudnn · a90c7e81
  由 sentient07 提交于 6月 09, 2016
  
  a90c7e81
- Added register_opt2 tags for other opts · 3c5c1506
  由 sentient07 提交于 6月 09, 2016
  
  3c5c1506
- Fix for SupportCodeError and KeyError · 9dfbff30
  由 sentient07 提交于 6月 07, 2016
  
  9dfbff30
- Changed op_lifter to accept Op and Inputs · 3f9d0601
  由 sentient07 提交于 6月 06, 2016
  
  3f9d0601
- Advance GraphToGPU. Make register_opt2 to register to it to bypass op_lifter · e45b6cd6
  由 Frederic Bastien 提交于 6月 02, 2016
  
  e45b6cd6
- GraphToGPUDB make it use the right tags · d9d4fc8b
  由 Frederic Bastien 提交于 6月 02, 2016
  
  d9d4fc8b
- Add constant to the mapping and don't add all outputs. · c796ef3b
  由 Frederic Bastien 提交于 6月 02, 2016
  
  c796ef3b